The SVB Bank Run and the USDC Crisis


by Ariah Klages-Mundt

7 of the largest 10 stablecoins depegged as a massive bank run effect rippled across crypto. What happened and what the lessons are for the space.

Starting Friday, March 11, and persisting through the weekend, most major stablecoins lost their peg and stablecoin liquidity virtually evaporated.

Bank runs and crypto runs

The trigger was a bank run on Silicon Valley Bank (SVB) and subsequent closing of the bank amidst illiquidity and insolvency worries. The intuition for a bank run is that, as a depositor, if the bank is insolvent (you don’t know) and you don’t rush for the exits, then you may be holding the bag for everyone who exits ahead of you. One brief caveat is that if you under the FDIC insurance limit, then your deposits are covered separately.

While stablecoins aren’t banks, they can experience runs too. Terra is a good example (and FTX, though not a stablecoin). A good article explaining bank runs and how they apply to crypto is here.

A key point is that a run on an insolvent institution is not the same as a run on a solvent one (aka a ‘Diamond-Dybvig’ bank run). An insolvent run is rational as there aren’t enough assets to go around, and depositors rush to cut their losses. On the other hand, a solvent run may be triggered or may not be. It is an irrational run in that no one has an incentive to run unless a run is underway anyway. The problem then arises from a liquidity mismatch: not all assets are liquid but, over time, assets can cover liabilities.

FTX was an insolvent run. But FTX wasn’t supposed to be a ‘bank’ with investments. It was supposed to be a custodian backed one-to-one (and communicated this to the public), which should be un-runnable. That it wasn’t is almost certainly fraud, which led to the insolvency.

Terra, although not a bank, was also analogous to an insolvent bank run. It was an insolvent-by-design protocol. In particular, its design paid out assets from new investors to earlier investors, leaving the protocol insolvent. The run that led to its collapse was rational, and also bears a lot of resemblance to currency attacks, like Soros’s famous attack on the British pound.

In the case of SVB, it appeared to be a rational run based on likely insolvency, which in turn triggered a run on USDC and spread the issue to crypto.


After Circle reported exposure to SVB for up to 25% of its reserves (later clarified down to 3.3b of ~40b), USDC holders discounted USDC’s value and rushed to redeem or sell it.

Redemptions were soon halted (reportedly for the weekend) when short-term constraints were reached. This led to more uncertainty about when redemptions might resume and sparked more heavy selling on secondary markets, driving USDC down to 0.87.


DAI largely mirrored USDC’s depeg below 0.90 Why? Maker’s chosen market design is to effectively peg to USDC instead of USD by allowing large amounts of 1:1 conversions. This led to ~1b of USDC being converted to DAI bringing prices in line.

This effect is despite a large portion of DAI (currently ~31%) being backed by other assets. The static coding of the 1:1 mechanism meant that governance was required to swiftly intervene to change the policy, but with decentralized governance doing this quickly is difficult.

DAI generated by collateral type as of Mar 11. (source: Dai Stats)

The problematic infrastructure was Maker’s Peg Stability Module (PSM). The 1:1 nature of the PSM meant that there was a run into the USDC PSM (using USDC to mint DAI) and a run out of the USDP PSM (redeeming DAI for USDP). Due to the design, traders could perform these swaps at non-market prices to the detriment of Maker and causing DAI to depeg.

This is a consequence of the market design that Maker chose, i.e., largely adopting USDC’s ‘primary market’ (by 1:1 hard-coding) and so suffering the same issues as USDC. We examined this issue here.

Static designs like the PSM mean that the stability of DAI in a crisis relies on governance to make quick corrective policy actions in a crisis. And decentralized governance is just not well equipped to this quickly.


Frax similarly depegged with USDC. In addition to being <100% reserved (it, like Terra was, is an insolvent-by-design protocol), Frax hardcodes price = $1 for all collateral assets: USDC, DAI, FEI, USDP, LUSD, sUSD, affecting redemption price. Manual governance intervention would have been needed to change this.


Liquity’s LUSD depegged to 0.97 but has recovered. While LUSD is independent of USDC, the crunch in overall liquidity likely caused a short-term sell out of LUSD. LUSD’s redemption mechanism provides price support that decays with current demand (as explained here), in line with the depeg and gradual recovery (as could be seen in historical data).

On-chain liquidity crunch

Another casualty of the crisis was on-chain stablecoin liquidity, which virtually dried up. This was an effect of the market structure of Curve: most liquidity evaporated as the USDC/DAI/USDT pool became ineffective. And the fact that major liquidity routes in Uniswap involved USDC.

Swaps between USDT and BUSD, two stablecoins less affected by the crisis, could only be done with high slippage due to the structure of on-chain liquidity.

Curve had $3.5b in assets, but the quality of liquidity relies primarily on the USDC/DAI/USDT 3-pool. With prices vastly different between all three assets, the Stableswap curve design simply can’t provide good liquidity for any pair in the pool. This issue extended to most other pools due to the fact that those pools composed their liquidity with the 3-pool (and so orders would have to route effectively through both pools).

Asset composition of the Curve 3-pool on Mar 10–11. The imbalance in USDC means the pool cannot offer good liquidity for any other pairs. (source: Dune dashboard)

Liquidity providers (LPs) in Uniswap v3 can theoretically adapt, but in practice they weren’t adapting. It was a dangerous time to LP as it wasn’t clear what many prices were or how to define good LP price ranges. Most major liquidity routes, like DAI/USDC and USDC/USDT, dried up.

Almost no Uniswap liquidity for DAI/USDC was available on Mar 11. The 0.05% fee pool was similar
Almost no Uniswap liquidity for USDC/USDT was available on Mar 11. The 0.05% fee pool was similar.

A few stablecoins remained close(r) to peg. USDT in particular, but also TUSD, BUSD, GUSD, and somewhat LUSD. But these are underdeveloped routes in Uniswap and their liquidity was either exhausted or used to go through USDC.

LUSD Uniswap liquidity was tied to USDC and dried up
USDP (aka PAX) liquidity was exhausted despite being paired with USDT

Peg recoveries

As more information came out about Circle’s precise exposure, it seemed there was a good chance for most pegs to recover. These stablecoins are not like Terra (aside from Frax) and are backed with real assets. It still remained to be seen what happened to Circle’s assets stuck at SVB, but the possible losses appear to be less than first worried.

One further worry was of contagion possibilities in the banking system, with possibly other banks becoming affects. And in turn possibly more effects on centralized stablecoins that hold reserves at such banks.

The first issue was resolved on the Sunday when the Federal Reserve announced that all depositors would be protected at SVB as well at Signature Bank. And the second issue was cleared up with the creation of the Bank Term Funding Program, which could lend cash to banks against par value of collateral as opposed to depressed market values (which is a different form of bail-out).

What can DeFi do better?

The crisis exposed cut corners, like:

  • Assuming prices = 1 for collateral stablecoins instead of using robust asset pricing.
  • Lack of diversification in on-chain reserve assets.
  • Lack of contingency mechanisms, circuit breakers.
  • Relying on gov interventions to adapt protocols to new market conditions.

The last two go hand-in-hand. If protocols don’t automate important contingency decisions, they must rely on slow-moving decentralized governance to maintain safety like in manually adapting the DAI PSM after the USDC depeg. We discuss this here:

Another benefit of automation is that the rules are clear to everyone, things can be much more transparent as well as more resilient. This brings us back to some of the core principles of DeFi, where it can really shine! We discuss this here:

How to fix the issues that arose this weekend and build more all-around resilient mechanisms? For some possible routes, check out the mechanisms that we designed for Gyroscope.